题名 | 基于 CUDA 的粒子模拟并行技术研究 |
作者 | 杜克伟 |
学位类别 | 硕士 |
答辩日期 | 2015-05-26 |
授予单位 | 中国科学院研究生院 |
授予地点 | 北京 |
导师 | 徐进 |
关键词 | CUDA 直线加速器 粒子模拟 PIC 泊松方程 |
学位专业 | 计算机应用技术 |
中文摘要 | 高能粒子直线加速器是采用沿直线轨道分布的高频电磁场加速电子、质子和其它重 离子的装置。根据加速粒子的不同分为电子直线加速器、质子加速器、重离子直线加速 器等。加速器不仅在科研领域有着广泛的应用,在医学上也用于癌症的放射治疗。近年 来超级计算机技术得到了突飞猛进的快速发展。与此同时,粒子加速模拟在设计和优化 粒子加速器中扮演着越来越重要的角色。随着超级计算机技术的快速发展,使得利用超 级计算机强大的计算能力进行超大规模粒子加速模拟成为可能。 利用超级计算机技术的优势,开发出能够充分利用计算资源模拟各种能量和高密度 的直线加速器模拟软件具有非常重要的意义。CUDA(Compute Unified Device Architec ture)是显卡厂商 NVIDIA 推出致力于通用大规模并行计算的基础平台。自推出以来便受 到从事高性能计算工作人员的追捧,经过十几年的不断更新,该平台日趋成熟和完善。该 平台可以使传统的图形处理器 GPU 能够解决复杂的大规模并行计算问题。基于该平台所 编写出的程序可以在支持 CUDA™的图形处理器上以超高性能运行。因此使用 CUDA 架构的 超强的并行计算能力为大规模粒子加速模拟提速是一个非常值得尝试和有意义的工作。 本文介绍了基于 CUDA 架构开发的三维异构并行束流动力学模拟软件 LOCUS3DG 的开 发进展。该软件是基于课题组以前开发的并行束流模拟软件 LOCUS3D,该软件采用双精 度数据类型,为了利用单精度数据并行通讯量小的特点,本论文研究了使用单精度和双 精度两种数据类型的情况,并研究了相应的泊松方程求解器在 CUDA 异构环境下的性能。 在国家超算天津中心的集群上对开发的算法和软件进行了详细的测试,得到了较好的测 试结果。论文最后总结了开发的异构束流模拟软件 LOCUS3DG 软件 的优势和现状及将来 进一步的研究展望。 |
英文摘要 | High-energy particle linear accelerator is a device that can accelerate electrons, proto ns or other heavy ions using linear high-frequency electromagnetic field along a straigh t track. According to the accelerated particles, it can be classified into electron linear accelerators, proton linear accelerators, heavy ion linear accelerator and so on. Not onl y in the field of scientific research been widely used, but it also been used in medicin e used for radiation therapy of cancer. In recent years, supercomputer technology has r apid development. Meanwhile, the particle acceleration simulation plays an increasingly important role in the design and optimization of particle accelerators. With the rapid development of supercomputer technology, ultra large-scale simulations of particle accel eration on supercomputers have come true. Making full use of supercomputer to developed simulation software for a variety of high-energy and high-density linear accelerator has very important significance. CUDA (Compute Unified Device Architecture) is launched by GPU manufacturers NVIDIA an d committed to the basis of general large-scale parallel computing platform. Since its l aunch, it has been engaged in the pursuit of high-performance computing, after years o f constantly update, the platform has become increasingly mature and powerful. The pl atform can make use of traditional GPU’scapability of massively parallel computing to solve complex problems. Based on this platform one can develop software that has hig h performance using CUDA ™ supported graphics processor. Therefore, the use of CU DA architecture for large-scale parallel computing to accelerate the simulation of particl e acceleration is worth to try and is a meaningful work. This article describes the development progress of three-dimensional heterogeneous p arallel beam dynamics simulation software LOCUS3DG based on CUDA architecture. T he software is based on parallel beam simulation software LOCUS3D,which is develo ped by our research group previously. It uses double precision data types. In order to take the advantage of less amount of parallel communication orf single-precision data, this thesis uses two data types, single-precision and double-precision, to study the corre sponding Poisson solver based on CUDA. On the National Supercomputer Center in Ti anjin, algorithms and software have been developedand detailed tests have been performed with good results. Finally, the thesis summarizes the advantages and status of the development of heterogeneous beam dynamics simulation software LOCUS3DG and gi ves further prospects in future research. |
语种 | 中文 |
学科主题 | 计算机仿真 |
公开日期 | 2015-06-23 |
内容类型 | 学位论文 |
源URL | [http://ir.iscas.ac.cn/handle/311060/17099] |
专题 | 软件研究所_并行计算实验室 _学位论文 |
推荐引用方式 GB/T 7714 | 杜克伟. 基于 CUDA 的粒子模拟并行技术研究[D]. 北京. 中国科学院研究生院. 2015. |
个性服务 |
查看访问统计 |
相关权益政策 |
暂无数据 |
收藏/分享 |
除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。
修改评论